Model Card Digest
Key Amos 2.8 specs at a glance; full architecture, training recipe, and intended use live on the Models page.
| Item | Value |
|---|---|
| Model name | amos-2.8 |
| Base model | Qwen3-8B (official pretrained weights, bf16, causal LM) |
| Total params | ≈ 8.2B (full base) + ≈ 0.35B trainable LoRA parameters |
| Effective context | max_len = 896 tokens (base limit 40,960) |
| Benchmarks | 4-round blind-test mean 97.45% ±0.6pp; strict OOD 96.05% leads jev-1.13 |
| Latency (measured) | ≈950–1200 ms (public e2e); 0 errors across 2,000 calls |